{"id":"f5dcccbc29f8ef22ce3ecd828303f7eb765c20d09c6f17d973a82bfa42af6430","title":"arXiv cs.AI authors: Clinical AI can reach the right diagnosis by an unsafe route","argument":"The authors argue that a correct final diagnosis does not show that a clinical AI system made safe decisions along the way. Their trajectory audit replays real admissions checkpoint by checkpoint, judging each test, treatment and disposition for potential harm and clinical validity against the evidence available at that point. They also describe an oversight audit that examines what happens after corrective feedback: whether it repairs an error and whether new risks follow.\n\nIn the studied cases, 55.2% of hospital trajectories that reached an acceptable final diagnosis included a decision with moderate or severe potential harm. Structured corrections resolved more first errors than bare rejection, but new potentially unsafe decisions followed both. The authors’ central warning is that neither a correct diagnosis nor a successful correction certifies a safe sequence of decisions. The reported rates come from the cases and systems in this study; they do not by themselves establish how often the same pattern occurs in other clinical settings.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The reported results cover 300 cases from a hospital cardiology cohort, MIMIC-IV and ASCENT, with up to seven AI systems per source; the abstract does not report results for other clinical settings. The abstract describes a physician reference standard based on triple annotation and senior adjudication of a stratified sample, but does not give the sample size or further calibration details.","speaker":{"name":"Chen, Guanqiao, Jiang, Wei, Zhang, Shu, Lin, Qian, Yu, Guanxu, Li, Siran, Wang, Chunhui, Shi, Tianyu","roleAtSource":"Article author"},"source":{"title":"Right Diagnosis, Unsafe Route: Auditing Clinical AI Decision Trajectories and Corrective Oversight for Validity and Potential Harm","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.22612","publishedAt":"2026-11-17","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors argue that a correct final diagnosis can coexist with potentially harmful intermediate decisions, and that evaluating only final diagnoses misses this risk.","excerpt":"A clinical AI system can reach a correct final diagnosis through intermediate decisions that could have harmed the patient","url":"https://arxiv.org/abs/2611.22612","relationship":"Source-stated"},{"text":"Among hospital trajectories reaching an acceptable final diagnosis, 55.2% passed through a decision with moderate or severe potential harm.","excerpt":"55.2% of hospital trajectories reaching an acceptable final diagnosis passed through a decision with moderate or severe potential harm","url":"https://arxiv.org/abs/2611.22612","relationship":"Source-stated"},{"text":"Structured corrections resolved more first errors than bare rejection, but new potentially unsafe decisions followed both kinds of feedback.","excerpt":"Structured corrections resolved 73.6% of first errors versus 24.8% after bare rejection, yet new potentially unsafe decisions followed 30.2% of corrections versus 19.7% of rejections.","url":"https://arxiv.org/abs/2611.22612","relationship":"Source-stated"},{"text":"The authors warn that repairing an error does not establish that the subsequent sequence of decisions is safe.","excerpt":"A correct final diagnosis does not certify a safe trajectory, and a successful correction does not certify a safe recovery.","url":"https://arxiv.org/abs/2611.22612","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.22612","checkedAt":"2026-11-18T16:50:12.878Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How do the reported rates change when the trajectory audit and oversight audit are applied to clinical cases and systems beyond those studied?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"arXiv cs.AI authors: Clinical AI can reach the right diagnosis by an unsafe route","summary":"The authors argue that a correct final diagnosis does not show that a clinical AI system made safe decisions along the way. Their trajectory audit replays real admissions checkpoint by checkpoint, judging each test, treatment and disposition for potential harm and clinical validity against the evidence available at that point. They also describe an oversight audit that examines what happens after corrective feedback: whether it repairs an error and whether new risks follow.\n\nIn the studied cases, 55.2% of hospital trajectories that reached an acceptable final diagnosis included a decision with moderate or severe potential harm. Structured corrections resolved more first errors than bare rejection, but new potentially unsafe decisions followed both. The authors’ central warning is that neither a correct diagnosis nor a successful correction certifies a safe sequence of decisions. The reported rates come from the cases and systems in this study; they do not by themselves establish how often the same pattern occurs in other clinical settings.","question":"How do the reported rates change when the trajectory audit and oversight audit are applied to clinical cases and systems beyond those studied?","claims":[{"text":"The authors argue that a correct final diagnosis can coexist with potentially harmful intermediate decisions, and that evaluating only final diagnoses misses this risk.","excerpt":"A clinical AI system can reach a correct final diagnosis through intermediate decisions that could have harmed the patient","support":"source-stated"},{"text":"Among hospital trajectories reaching an acceptable final diagnosis, 55.2% passed through a decision with moderate or severe potential harm.","excerpt":"55.2% of hospital trajectories reaching an acceptable final diagnosis passed through a decision with moderate or severe potential harm","support":"source-stated"},{"text":"Structured corrections resolved more first errors than bare rejection, but new potentially unsafe decisions followed both kinds of feedback.","excerpt":"Structured corrections resolved 73.6% of first errors versus 24.8% after bare rejection, yet new potentially unsafe decisions followed 30.2% of corrections versus 19.7% of rejections.","support":"source-stated"},{"text":"The authors warn that repairing an error does not establish that the subsequent sequence of decisions is safe.","excerpt":"A correct final diagnosis does not certify a safe trajectory, and a successful correction does not certify a safe recovery.","support":"source-stated"}],"limitations":["The reported results cover 300 cases from a hospital cardiology cohort, MIMIC-IV and ASCENT, with up to seven AI systems per source; the abstract does not report results for other clinical settings.","The abstract describes a physician reference standard based on triple annotation and senior adjudication of a stratified sample, but does not give the sample size or further calibration details."],"mediaId":"b01dd6abddad098290aaa31fb73cc3fa7a1fd7e850831f31f8ee96fc8f402f9e","correctionReason":null,"id":"f5dcccbc29f8ef22ce3ecd828303f7eb765c20d09c6f17d973a82bfa42af6430","publicationId":"f1516b39-3521-4027-8d25-f35bd528a5a8","version":1,"sourceUrl":"https://arxiv.org/abs/2611.22612","sourceTitle":"Right Diagnosis, Unsafe Route: Auditing Clinical AI Decision Trajectories and Corrective Oversight for Validity and Potential Harm","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-17","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-18T05:13:12.469Z","createdAt":"2026-11-18T16:34:29.297Z","historical":false,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"b01dd6abddad098290aaa31fb73cc3fa7a1fd7e850831f31f8ee96fc8f402f9e","kind":"article","url":"https://arxiv.org/abs/2611.22612","sourceUrl":"https://arxiv.org/abs/2611.22612","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-18T16:33:28.469Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.22612","checkedAt":"2026-11-18T16:50:12.878Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v9","mediaId":"b01dd6abddad098290aaa31fb73cc3fa7a1fd7e850831f31f8ee96fc8f402f9e","confidence":0.87,"provider":"typesafe","model":"jev-1.13.0","inputHash":"675fe740305c7070bb331231e2eadbbceea02af0c35e482cce7741d796ac8e41","decidedAt":"2026-11-18T16:33:31.673Z"},"provenance":{"textHash":"03157f98b0321236b9e78c487c28ea85f22eedf73511b961ad28f181871839dc","observationId":"726e9b1a57d138d4acefb74ac5412e0c75c1b2820514482de9f4a1e419ffe469","retrievedAt":"2026-11-18T16:33:28.116Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"4eabd287-9218-4219-8917-c5a613e5cf75"},"reviewer":{"model":"gpt-6-luna","runId":"b75c63fa-a442-4ad1-866d-436d9f54d9c7"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-18T16:34:29.297Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Chen, Guanqiao","url":null,"kind":"unknown"},{"name":"Jiang, Wei","url":null,"kind":"unknown"},{"name":"Zhang, Shu","url":null,"kind":"unknown"},{"name":"Lin, Qian","url":null,"kind":"unknown"},{"name":"Yu, Guanxu","url":null,"kind":"unknown"},{"name":"Li, Siran","url":null,"kind":"unknown"},{"name":"Wang, Chunhui","url":null,"kind":"unknown"},{"name":"Shi, Tianyu","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Chen, Guanqiao","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Chen, Guanqiao","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.22612","locator":"head/meta[name=author|citation_author]"},"assertionId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-18T16:50:12.878Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac","sourceUrl":"https://arxiv.org/abs/2611.22612","contentSha256":"85ce8e2ff4fbfddddf4c8a6b72479a32f0c1d93ec976f5ed0940bb4577b41332","textHash":"03157f98b0321236b9e78c487c28ea85f22eedf73511b961ad28f181871839dc","observedAt":"2026-11-18T16:50:12.878Z"}},{"actor":{"name":"Jiang, Wei","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Jiang, Wei","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.22612","locator":"head/meta[name=author|citation_author]"},"assertionId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-18T16:50:12.878Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac","sourceUrl":"https://arxiv.org/abs/2611.22612","contentSha256":"85ce8e2ff4fbfddddf4c8a6b72479a32f0c1d93ec976f5ed0940bb4577b41332","textHash":"03157f98b0321236b9e78c487c28ea85f22eedf73511b961ad28f181871839dc","observedAt":"2026-11-18T16:50:12.878Z"}},{"actor":{"name":"Zhang, Shu","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Zhang, Shu","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.22612","locator":"head/meta[name=author|citation_author]"},"assertionId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-18T16:50:12.878Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac","sourceUrl":"https://arxiv.org/abs/2611.22612","contentSha256":"85ce8e2ff4fbfddddf4c8a6b72479a32f0c1d93ec976f5ed0940bb4577b41332","textHash":"03157f98b0321236b9e78c487c28ea85f22eedf73511b961ad28f181871839dc","observedAt":"2026-11-18T16:50:12.878Z"}},{"actor":{"name":"Lin, Qian","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Lin, Qian","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.22612","locator":"head/meta[name=author|citation_author]"},"assertionId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-18T16:50:12.878Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac","sourceUrl":"https://arxiv.org/abs/2611.22612","contentSha256":"85ce8e2ff4fbfddddf4c8a6b72479a32f0c1d93ec976f5ed0940bb4577b41332","textHash":"03157f98b0321236b9e78c487c28ea85f22eedf73511b961ad28f181871839dc","observedAt":"2026-11-18T16:50:12.878Z"}},{"actor":{"name":"Yu, Guanxu","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Yu, Guanxu","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.22612","locator":"head/meta[name=author|citation_author]"},"assertionId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac:participation:4","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-18T16:50:12.878Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac","sourceUrl":"https://arxiv.org/abs/2611.22612","contentSha256":"85ce8e2ff4fbfddddf4c8a6b72479a32f0c1d93ec976f5ed0940bb4577b41332","textHash":"03157f98b0321236b9e78c487c28ea85f22eedf73511b961ad28f181871839dc","observedAt":"2026-11-18T16:50:12.878Z"}},{"actor":{"name":"Li, Siran","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Li, Siran","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.22612","locator":"head/meta[name=author|citation_author]"},"assertionId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac:participation:5","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-18T16:50:12.878Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac","sourceUrl":"https://arxiv.org/abs/2611.22612","contentSha256":"85ce8e2ff4fbfddddf4c8a6b72479a32f0c1d93ec976f5ed0940bb4577b41332","textHash":"03157f98b0321236b9e78c487c28ea85f22eedf73511b961ad28f181871839dc","observedAt":"2026-11-18T16:50:12.878Z"}},{"actor":{"name":"Wang, Chunhui","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Wang, Chunhui","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.22612","locator":"head/meta[name=author|citation_author]"},"assertionId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac:participation:6","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-18T16:50:12.878Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac","sourceUrl":"https://arxiv.org/abs/2611.22612","contentSha256":"85ce8e2ff4fbfddddf4c8a6b72479a32f0c1d93ec976f5ed0940bb4577b41332","textHash":"03157f98b0321236b9e78c487c28ea85f22eedf73511b961ad28f181871839dc","observedAt":"2026-11-18T16:50:12.878Z"}},{"actor":{"name":"Shi, Tianyu","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Shi, Tianyu","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.22612","locator":"head/meta[name=author|citation_author]"},"assertionId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac:participation:7","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-18T16:50:12.878Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac","sourceUrl":"https://arxiv.org/abs/2611.22612","contentSha256":"85ce8e2ff4fbfddddf4c8a6b72479a32f0c1d93ec976f5ed0940bb4577b41332","textHash":"03157f98b0321236b9e78c487c28ea85f22eedf73511b961ad28f181871839dc","observedAt":"2026-11-18T16:50:12.878Z"}}],"participationCoverage":{"total":8,"returned":8,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.22612","snapshotId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac","contentSha256":"85ce8e2ff4fbfddddf4c8a6b72479a32f0c1d93ec976f5ed0940bb4577b41332","textHash":"03157f98b0321236b9e78c487c28ea85f22eedf73511b961ad28f181871839dc","observedAt":"2026-11-18T16:50:12.878Z","basis":"publisher-page-metadata","sourceLinks":46,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/f5dcccbc29f8ef22ce3ecd828303f7eb765c20d09c6f17d973a82bfa42af6430/source"}},"canonicalUrl":"https://escapemeridian.com/readings/arxiv-cs-ai-authors-clinical-ai-can-reach-the-right-diagnosis-f5dcccbc","formats":{"html":"/readings/arxiv-cs-ai-authors-clinical-ai-can-reach-the-right-diagnosis-f5dcccbc","markdown":"/readings/arxiv-cs-ai-authors-clinical-ai-can-reach-the-right-diagnosis-f5dcccbc.md","json":"/readings/arxiv-cs-ai-authors-clinical-ai-can-reach-the-right-diagnosis-f5dcccbc.json"},"attribution":"Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.","maintainedGuideLinks":[],"conditions":[],"readerContext":{"version":"reader-context-v1","attribution":"Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.","terms":[],"entities":[{"id":"arxiv","name":"arXiv","kind":"organization","context":"arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.","referenceUrls":["https://arxiv.org/","https://info.arxiv.org/help/api/index.html"],"path":"/entities/arxiv","json":"/entities/arxiv.json","markdown":"/entities/arxiv.md"}]},"researchQuestions":[],"withdrawn":false,"references":{"schemaVersion":1,"target":{"type":"reading","id":"f5dcccbc29f8ef22ce3ecd828303f7eb765c20d09c6f17d973a82bfa42af6430","revision":"f1516b39-3521-4027-8d25-f35bd528a5a8","revisionKind":"publication-id","title":"arXiv cs.AI authors: Clinical AI can reach the right diagnosis by an unsafe route","url":"https://escapemeridian.com/readings/arxiv-cs-ai-authors-clinical-ai-can-reach-the-right-diagnosis-f5dcccbc"},"source":{"type":"source","id":"https://arxiv.org/abs/2611.22612","revision":"03157f98b0321236b9e78c487c28ea85f22eedf73511b961ad28f181871839dc","revisionKind":"sha256-source-text","title":"Right Diagnosis, Unsafe Route: Auditing Clinical AI Decision Trajectories and Corrective Oversight for Validity and Potential Harm","url":"https://arxiv.org/abs/2611.22612"},"status":"available","items":[],"next":null,"scope":"Explicit citations in public readings to this reading. Shared topics are not citations. Other record types need an explicit reference adapter."},"revisions":[{"id":"f5dcccbc29f8ef22ce3ecd828303f7eb765c20d09c6f17d973a82bfa42af6430","title":"arXiv cs.AI authors: Clinical AI can reach the right diagnosis by an unsafe route","argument":"The authors argue that a correct final diagnosis does not show that a clinical AI system made safe decisions along the way. Their trajectory audit replays real admissions checkpoint by checkpoint, judging each test, treatment and disposition for potential harm and clinical validity against the evidence available at that point. They also describe an oversight audit that examines what happens after corrective feedback: whether it repairs an error and whether new risks follow.\n\nIn the studied cases, 55.2% of hospital trajectories that reached an acceptable final diagnosis included a decision with moderate or severe potential harm. Structured corrections resolved more first errors than bare rejection, but new potentially unsafe decisions followed both. The authors’ central warning is that neither a correct diagnosis nor a successful correction certifies a safe sequence of decisions. The reported rates come from the cases and systems in this study; they do not by themselves establish how often the same pattern occurs in other clinical settings.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The reported results cover 300 cases from a hospital cardiology cohort, MIMIC-IV and ASCENT, with up to seven AI systems per source; the abstract does not report results for other clinical settings. The abstract describes a physician reference standard based on triple annotation and senior adjudication of a stratified sample, but does not give the sample size or further calibration details.","speaker":{"name":"Chen, Guanqiao, Jiang, Wei, Zhang, Shu, Lin, Qian, Yu, Guanxu, Li, Siran, Wang, Chunhui, Shi, Tianyu","roleAtSource":"Article author"},"source":{"title":"Right Diagnosis, Unsafe Route: Auditing Clinical AI Decision Trajectories and Corrective Oversight for Validity and Potential Harm","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.22612","publishedAt":"2026-11-17","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors argue that a correct final diagnosis can coexist with potentially harmful intermediate decisions, and that evaluating only final diagnoses misses this risk.","excerpt":"A clinical AI system can reach a correct final diagnosis through intermediate decisions that could have harmed the patient","url":"https://arxiv.org/abs/2611.22612","relationship":"Source-stated"},{"text":"Among hospital trajectories reaching an acceptable final diagnosis, 55.2% passed through a decision with moderate or severe potential harm.","excerpt":"55.2% of hospital trajectories reaching an acceptable final diagnosis passed through a decision with moderate or severe potential harm","url":"https://arxiv.org/abs/2611.22612","relationship":"Source-stated"},{"text":"Structured corrections resolved more first errors than bare rejection, but new potentially unsafe decisions followed both kinds of feedback.","excerpt":"Structured corrections resolved 73.6% of first errors versus 24.8% after bare rejection, yet new potentially unsafe decisions followed 30.2% of corrections versus 19.7% of rejections.","url":"https://arxiv.org/abs/2611.22612","relationship":"Source-stated"},{"text":"The authors warn that repairing an error does not establish that the subsequent sequence of decisions is safe.","excerpt":"A correct final diagnosis does not certify a safe trajectory, and a successful correction does not certify a safe recovery.","url":"https://arxiv.org/abs/2611.22612","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.22612","checkedAt":"2026-11-18T16:50:12.878Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How do the reported rates change when the trajectory audit and oversight audit are applied to clinical cases and systems beyond those studied?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"arXiv cs.AI authors: Clinical AI can reach the right diagnosis by an unsafe route","summary":"The authors argue that a correct final diagnosis does not show that a clinical AI system made safe decisions along the way. Their trajectory audit replays real admissions checkpoint by checkpoint, judging each test, treatment and disposition for potential harm and clinical validity against the evidence available at that point. They also describe an oversight audit that examines what happens after corrective feedback: whether it repairs an error and whether new risks follow.\n\nIn the studied cases, 55.2% of hospital trajectories that reached an acceptable final diagnosis included a decision with moderate or severe potential harm. Structured corrections resolved more first errors than bare rejection, but new potentially unsafe decisions followed both. The authors’ central warning is that neither a correct diagnosis nor a successful correction certifies a safe sequence of decisions. The reported rates come from the cases and systems in this study; they do not by themselves establish how often the same pattern occurs in other clinical settings.","question":"How do the reported rates change when the trajectory audit and oversight audit are applied to clinical cases and systems beyond those studied?","claims":[{"text":"The authors argue that a correct final diagnosis can coexist with potentially harmful intermediate decisions, and that evaluating only final diagnoses misses this risk.","excerpt":"A clinical AI system can reach a correct final diagnosis through intermediate decisions that could have harmed the patient","support":"source-stated"},{"text":"Among hospital trajectories reaching an acceptable final diagnosis, 55.2% passed through a decision with moderate or severe potential harm.","excerpt":"55.2% of hospital trajectories reaching an acceptable final diagnosis passed through a decision with moderate or severe potential harm","support":"source-stated"},{"text":"Structured corrections resolved more first errors than bare rejection, but new potentially unsafe decisions followed both kinds of feedback.","excerpt":"Structured corrections resolved 73.6% of first errors versus 24.8% after bare rejection, yet new potentially unsafe decisions followed 30.2% of corrections versus 19.7% of rejections.","support":"source-stated"},{"text":"The authors warn that repairing an error does not establish that the subsequent sequence of decisions is safe.","excerpt":"A correct final diagnosis does not certify a safe trajectory, and a successful correction does not certify a safe recovery.","support":"source-stated"}],"limitations":["The reported results cover 300 cases from a hospital cardiology cohort, MIMIC-IV and ASCENT, with up to seven AI systems per source; the abstract does not report results for other clinical settings.","The abstract describes a physician reference standard based on triple annotation and senior adjudication of a stratified sample, but does not give the sample size or further calibration details."],"mediaId":"b01dd6abddad098290aaa31fb73cc3fa7a1fd7e850831f31f8ee96fc8f402f9e","correctionReason":null,"id":"f5dcccbc29f8ef22ce3ecd828303f7eb765c20d09c6f17d973a82bfa42af6430","publicationId":"f1516b39-3521-4027-8d25-f35bd528a5a8","version":1,"sourceUrl":"https://arxiv.org/abs/2611.22612","sourceTitle":"Right Diagnosis, Unsafe Route: Auditing Clinical AI Decision Trajectories and Corrective Oversight for Validity and Potential Harm","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-17","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-18T05:13:12.469Z","createdAt":"2026-11-18T16:34:29.297Z","historical":false,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"b01dd6abddad098290aaa31fb73cc3fa7a1fd7e850831f31f8ee96fc8f402f9e","kind":"article","url":"https://arxiv.org/abs/2611.22612","sourceUrl":"https://arxiv.org/abs/2611.22612","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-18T16:33:28.469Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.22612","checkedAt":"2026-11-18T16:50:12.878Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v9","mediaId":"b01dd6abddad098290aaa31fb73cc3fa7a1fd7e850831f31f8ee96fc8f402f9e","confidence":0.87,"provider":"typesafe","model":"jev-1.13.0","inputHash":"675fe740305c7070bb331231e2eadbbceea02af0c35e482cce7741d796ac8e41","decidedAt":"2026-11-18T16:33:31.673Z"},"provenance":{"textHash":"03157f98b0321236b9e78c487c28ea85f22eedf73511b961ad28f181871839dc","observationId":"726e9b1a57d138d4acefb74ac5412e0c75c1b2820514482de9f4a1e419ffe469","retrievedAt":"2026-11-18T16:33:28.116Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"4eabd287-9218-4219-8917-c5a613e5cf75"},"reviewer":{"model":"gpt-6-luna","runId":"b75c63fa-a442-4ad1-866d-436d9f54d9c7"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-18T16:34:29.297Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Chen, Guanqiao","url":null,"kind":"unknown"},{"name":"Jiang, Wei","url":null,"kind":"unknown"},{"name":"Zhang, Shu","url":null,"kind":"unknown"},{"name":"Lin, Qian","url":null,"kind":"unknown"},{"name":"Yu, Guanxu","url":null,"kind":"unknown"},{"name":"Li, Siran","url":null,"kind":"unknown"},{"name":"Wang, Chunhui","url":null,"kind":"unknown"},{"name":"Shi, Tianyu","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Chen, Guanqiao","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Chen, Guanqiao","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.22612","locator":"head/meta[name=author|citation_author]"},"assertionId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-18T16:50:12.878Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac","sourceUrl":"https://arxiv.org/abs/2611.22612","contentSha256":"85ce8e2ff4fbfddddf4c8a6b72479a32f0c1d93ec976f5ed0940bb4577b41332","textHash":"03157f98b0321236b9e78c487c28ea85f22eedf73511b961ad28f181871839dc","observedAt":"2026-11-18T16:50:12.878Z"}},{"actor":{"name":"Jiang, Wei","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Jiang, Wei","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.22612","locator":"head/meta[name=author|citation_author]"},"assertionId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-18T16:50:12.878Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac","sourceUrl":"https://arxiv.org/abs/2611.22612","contentSha256":"85ce8e2ff4fbfddddf4c8a6b72479a32f0c1d93ec976f5ed0940bb4577b41332","textHash":"03157f98b0321236b9e78c487c28ea85f22eedf73511b961ad28f181871839dc","observedAt":"2026-11-18T16:50:12.878Z"}},{"actor":{"name":"Zhang, Shu","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Zhang, Shu","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.22612","locator":"head/meta[name=author|citation_author]"},"assertionId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-18T16:50:12.878Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac","sourceUrl":"https://arxiv.org/abs/2611.22612","contentSha256":"85ce8e2ff4fbfddddf4c8a6b72479a32f0c1d93ec976f5ed0940bb4577b41332","textHash":"03157f98b0321236b9e78c487c28ea85f22eedf73511b961ad28f181871839dc","observedAt":"2026-11-18T16:50:12.878Z"}},{"actor":{"name":"Lin, Qian","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Lin, Qian","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.22612","locator":"head/meta[name=author|citation_author]"},"assertionId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-18T16:50:12.878Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac","sourceUrl":"https://arxiv.org/abs/2611.22612","contentSha256":"85ce8e2ff4fbfddddf4c8a6b72479a32f0c1d93ec976f5ed0940bb4577b41332","textHash":"03157f98b0321236b9e78c487c28ea85f22eedf73511b961ad28f181871839dc","observedAt":"2026-11-18T16:50:12.878Z"}},{"actor":{"name":"Yu, Guanxu","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Yu, Guanxu","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.22612","locator":"head/meta[name=author|citation_author]"},"assertionId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac:participation:4","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-18T16:50:12.878Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac","sourceUrl":"https://arxiv.org/abs/2611.22612","contentSha256":"85ce8e2ff4fbfddddf4c8a6b72479a32f0c1d93ec976f5ed0940bb4577b41332","textHash":"03157f98b0321236b9e78c487c28ea85f22eedf73511b961ad28f181871839dc","observedAt":"2026-11-18T16:50:12.878Z"}},{"actor":{"name":"Li, Siran","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Li, Siran","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.22612","locator":"head/meta[name=author|citation_author]"},"assertionId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac:participation:5","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-18T16:50:12.878Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac","sourceUrl":"https://arxiv.org/abs/2611.22612","contentSha256":"85ce8e2ff4fbfddddf4c8a6b72479a32f0c1d93ec976f5ed0940bb4577b41332","textHash":"03157f98b0321236b9e78c487c28ea85f22eedf73511b961ad28f181871839dc","observedAt":"2026-11-18T16:50:12.878Z"}},{"actor":{"name":"Wang, Chunhui","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Wang, Chunhui","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.22612","locator":"head/meta[name=author|citation_author]"},"assertionId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac:participation:6","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-18T16:50:12.878Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac","sourceUrl":"https://arxiv.org/abs/2611.22612","contentSha256":"85ce8e2ff4fbfddddf4c8a6b72479a32f0c1d93ec976f5ed0940bb4577b41332","textHash":"03157f98b0321236b9e78c487c28ea85f22eedf73511b961ad28f181871839dc","observedAt":"2026-11-18T16:50:12.878Z"}},{"actor":{"name":"Shi, Tianyu","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Shi, Tianyu","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.22612","locator":"head/meta[name=author|citation_author]"},"assertionId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac:participation:7","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-18T16:50:12.878Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac","sourceUrl":"https://arxiv.org/abs/2611.22612","contentSha256":"85ce8e2ff4fbfddddf4c8a6b72479a32f0c1d93ec976f5ed0940bb4577b41332","textHash":"03157f98b0321236b9e78c487c28ea85f22eedf73511b961ad28f181871839dc","observedAt":"2026-11-18T16:50:12.878Z"}}],"participationCoverage":{"total":8,"returned":8,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.22612","snapshotId":"d1b54c4e6ea405f0d73be24290fa8c1546d01af26517ed7aec0a543b2bc166ac","contentSha256":"85ce8e2ff4fbfddddf4c8a6b72479a32f0c1d93ec976f5ed0940bb4577b41332","textHash":"03157f98b0321236b9e78c487c28ea85f22eedf73511b961ad28f181871839dc","observedAt":"2026-11-18T16:50:12.878Z","basis":"publisher-page-metadata","sourceLinks":46,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/f5dcccbc29f8ef22ce3ecd828303f7eb765c20d09c6f17d973a82bfa42af6430/source"}}}],"audits":[{"publicationId":"f1516b39-3521-4027-8d25-f35bd528a5a8","verdict":"uncertain","findings":[{"dimension":"attribution","status":"pass","explanation":"The argument and findings are attributed to the paper's authors, and the source lists the author names. The label 'arXiv cs.AI authors' is vague but not wrong. The draft keeps the results as the authors' reported findings.","excerpt":"Authors:Guanqiao Chen, Wei Jiang, Shu Zhang, Qian Lin, Guanxu Yu, Siran Li, Chunhui Wang, Tianyu Shi"},{"dimension":"summary","status":"uncertain","explanation":"Two omissions affect interpretation. First, the draft says new unsafe decisions 'followed both' kinds of feedback, which hides the abstract's trade-off: structured corrections fixed more errors but were followed by new unsafe decisions more often (30.2% vs 19.7%). Second, it omits that an LLM judge scored every decision, with physician review calibrating the rates. Smallest repair: add both rates and the LLM judge. Central terms such as 'trajectory' and 'disposition' (a patient's admit, discharge or transfer decision) also need plain-language explanation or tooltips. The 76.2% and 39.6% harm rates on the other datasets are absent from the summary.","excerpt":"yet new potentially unsafe decisions followed 30.2% of corrections versus 19.7% of rejections"},{"dimension":"date","status":"pass","explanation":"The originalPublished date of 2026-11-17 matches the arXiv v1 submission date. It is not the later feed date.","excerpt":"[Submitted on 17 Nov 2026]"},{"dimension":"support","status":"pass","explanation":"Each claim matches the abstract. The limitations accurately describe the case scope and the unspecified size of the physician sample. The question is a non-leading, open generalization issue. No unsupported benefits or costs are implied.","excerpt":"55.2% of hospital trajectories reaching an acceptable final diagnosis passed through a decision with moderate or severe potential harm"},{"dimension":"media","status":"pass","explanation":"The selected media is the verified link to the original arXiv abstract page. It is the direct source.","excerpt":null}],"createdAt":"2026-11-18T16:34:29.297Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}